The data was gathered from diverse sources of MIDI compositions and formatted for use in Language Model training.